Department of Mathematics and Statistics

Research Projects

Resource

2018 REU HPC training: [link]

Matlab at TACC: [link]

Matlab at PSC: [link]

Matlab at SDSC: [link]

2018 Projects

Potential projects to be considered:

  • Face recognition with Morph-II database.
  • Linear or Nonlinear Dimension Reduction techniques.
  • Age estimation with Morph-II database.
  • Gender/race classification with Morph-II database.
  • Preliminary study with Deep Learning.

2017 Projects

Database for computer vision and pattern recognition

The 2008 version of non-commercial release of the MORPH-II dataset will be used and referred to as MORPH-II. This dataset is a collection of 55,134 mugshots for 13,617 subjects taken between 2003 and late 2007, and includes many images of individuals that were arrested multiple times over this 5 year span. This gives the data a longitudinal aspect that has made it very useful in the field of computer vision and pattern recognition. In particular, the MORPH-II dataset is widely utilized in research on gender and race classification, as well as age estimation and synthesis. As a longitudinal dataset, there are repeat subjects with multiple entries in the MORPH-II dataset.

Face Recognition

Morph-II was applied to various face recognition projects with feature extraction of 2DPCA, 2DLDA, Local Binary Patterm (LBP), Bio-Inspiring Feature (BIF). For example, a novel technique named random subspace two-dimensional LDA (RS-2DLDA) is developed for face recognition. This approach offers a number of improvements over the random subspace two-dimensional PCA (RS- 2DPCA) framework introduced by Nguyen et al. [5]. Firstly, the eigenvectors from 2DLDA have more discriminative power than those from 2DPCA, resulting in higher accuracy for the RS-2DLDA method over RS-2DPCA. Various distance metrics are evaluated, and a weighting scheme is developed to further boost accuracy. A series of experiments on the MORPH-II and ORL datasets are conducted to demonstrate the effectiveness of this approach.

Gender Classification

We investigated a comparative analysis of feature extraction and dimension reduction methods used to classify race on a large data set. The feature extraction methods considered are Local Binary Patterns (LBP), Histograms of Oriented Gradients (HOG) and Bio-Inspired Features (BIF). The dimension reduction methods considered are Principle Component Analysis and Linear Discriminant Analysis. The experimental results based on a subset of the Morph-II dataset.

Age Estimation

We considered a global framework for age estimation that utilizes posterior probabilities from the race classification step to compute a race-composite age estimate. The first half of the paper is dedicated to summarizing the cleaning and image pre-processing stages of the proposed pipeline as they were applied to the MORPH-II dataset. In the second half, there is a more in-depth discussion of the novel components of this paper, including methodology, experiments, and some preliminary results.

Nonlinear Dimension Reduction Using Kernel methods

The following nonlinear dimensionality reduction techniques involving kernel functions are studied: kernel principal component analysis (KPCA), supervised kernel principal component analysis (SKPCA), and kernel Fisher's discriminant analysis (KFDA). These techniques are studied on 3 simulated datasets. Then a novel machine learning pipeline is proposed for the longitudinal face aging database MORPH-II. First, images in MORPH-II are preprocessed, then a number of features are extracted from the processed images. The KPCA, SKPCA, and KFDA techniques are used to transform then reduce the dimension of
the extracted features. The reduced dimension data serves as input for a linear support vector machine (SVM) to classify gender. Finally, the dimensionality reduction techniques are compared in terms of gender class cation results.

Disclaimer: 

This material is based upon work supported by the National Science Foundation under Grant Number (NSF DMS-1659288). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.